Methodology

We adopt PRITES (Huang & Lam, 2025, Monash Addiction Research Centre) as the governing frame for the web-scraped dataset: Provenance, Representativeness, Integrity, Transparency, Ethics, Sustainability.

• Provenance — every record stores its retrieval URL, fetch time, HTTP status, and page title.
• Representativeness — the discovery query bank spans 20+ languages across six continents (German, French, Spanish, Italian, Dutch, Polish, Czech, Hungarian, Swedish, Norwegian, Portuguese, Russian, Japanese, Korean, Mandarin, Hindi, Indonesian, Vietnamese, Thai, Arabic, Turkish, Hebrew and more), rotated deterministically by ISO week so every query runs once per full cycle rather than being sampled at random; registry, grant and directory channels run alongside it so that non-marketing, pre-commercial organisations are reachable at all.
• Integrity — every cited URL is re-fetched and required to actually mention the company.
• Transparency — taxonomy, scoring rubric, and source-tier rules are public on this page.
• Ethics — only public pages, identified bot UA, no scraping behind logins or paywalls.
• Sustainability — the validator re-runs on a schedule; flagged sources never silently disappear.

Huang, C. A., & Lam, T. (2025). PRITES: An integrative framework for investigating and assessing web-scraped HTTP-response datasets for research applications. arXiv:2511.13773.

Company discovery follows the four PRISMA-ScR stages, instrumented as pipeline counters:

1. Identification — candidate names emerge from seven parallel channels: the multilingual web sweep, recency-weighted news, ClinicalTrials.gov, ISRCTN / WHO ICTRP, NIH RePORTER SBIR/STTR awards, the Y Combinator directory, and openFDA device registers, plus manual passes over MInD, ScanMedicine, BiomedTracker, EU CTIS, MHRA PARD, PMDA, EMA, BfArM DiGA, CORDIS and Candid. The full inventory and strings are in §03 and §04.
2. Screening — sponsor and applicant names are filtered against an institution pattern (universities, hospitals, ministries) and a non-company host list, then the research agent retrieves each survivor and rejects out-of-scope names with a structured NOT_RELEVANT verdict. Every candidate ever seen — accepted or rejected — is logged with its verdict, so the funnel is auditable and rejected names are never re-researched.
3. Eligibility — surviving records must have a defensible substance / behavioural / recovery focus; pure mental-health and adjacent wellness apps are excluded, and borderline organisations are admitted but tagged adjacent with a written rationale rather than silently dropped.
4. Inclusion — the record is written with a confidence score and an immediate source-validation pass.

Tricco, A. C., et al. (2018). PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Annals of Internal Medicine, 169(7), 467–473.

A company atlas built on web search alone inherits the biases of search: it finds well-marketed, English-language, consumer-facing firms and misses pre-commercial, grant-funded, clinical-stage and non-Anglophone ones. The pipeline therefore runs several structurally different channels in parallel. The inventory below is imported by this page from the same module the pipeline runs on, so what is published is what executes.

The database architecture follows the NIHR Innovation Observatory horizon scan of healthcare technologies in development for addiction, extended with company-level sources — Y Combinator, NIH RePORTER, CORDIS and Candid — so that public and philanthropic capital is visible alongside venture capital.

automated — queried by the pipeline every cycle · site-restricted — reached through domain-limited search because the database has no open API · manual — subscription or non-machine-readable, run by hand on the strings below and entered through the admin console.

Wilkins, S., et al. (2025). A Pipeline Analysis of Healthcare Technologies in Development for Addiction. NIHR Innovation Observatory. doi:10.6084/m9.figshare.28771112.

Scope in one sentence: any organisation developing or delivering a product, service or technology for substance addiction (opioids and opiates, stimulants, cannabis and cannabinoids, alcohol, nicotine and tobacco, gabapentinoids, benzodiazepines, GHB, ketamine, novel psychoactive substances) or behavioural addiction (gambling, gaming, internet and smartphone, pornography, food), across treatment, medication, devices, digital therapeutics, harm reduction, recovery support and care navigation.

One deliberate divergence from the NIHR scan: that review excluded prevention, overdose response, nicotine and gambling to keep a clean treatment-pipeline frame. The Atlas includes all four, because the object here is an innovation ecosystem rather than a drug pipeline — and because nicotine and gambling are where much of the private capital sits. Preclinical-only research groups remain excluded.

Grey-literature items (company sites, press releases, aggregator entries) are appraised per source against the AACODS checklist — Authority, Accuracy, Coverage, Objectivity, Date, Significance. The validator currently operationalises three axes automatically; the others remain editorial.

Tyndall, J. (2010). AACODS Checklist. Flinders University.

The Atlas treats sources as a hierarchy. A green ✓ on a record is meaningful only when paired with the tier of the verifying source:

• Regulator — FDA, EMA, MHRA, BfArM, PMDA, NICE, SAMHSA, NIH, WHO, gov.uk, ec.europa.eu.
• Trial registry — ClinicalTrials.gov, EU CTR, ISRCTN, ANZCTR, DRKS, WHO ICTRP.
• Peer-reviewed — PubMed, OpenAlex, NEJM, Lancet, BMJ, JMIR, Nature, PLOS, Frontiers, CORDIS, Cochrane.
• Primary — the company's own domain (verified host match).
• News — recognised trade and financial press.
• Aggregator — Crunchbase, Pitchbook, Dealroom, LinkedIn, Wikipedia — useful for triangulation, never sufficient alone.

• URL returns HTTP ≥ 400, times out, or fails TLS.
• URL loads but does not mention the company or any alias.
• A record's only verified sources are aggregators or unknown-tier pages.

Flagged items are not deleted — they are displayed inline on the company profile with the failure reason, so reviewers can override or replace them.

Every company carries four subscores on a 0–100 scale and one composite. Nothing is hand-assigned: each subscore is a deterministic function of the record's own fields, recomputed on every ingest and every backfill, so two records with identical fields always receive identical scores. The rules below are the complete specification.

Composite

monitoring_priority = 0.25·innovation + 0.25·clinical + 0.25·regulatory + 0.25·commercial, rounded and clamped to 0–100. Equal weights are preregistered (OSF v1.0, §6.1) precisely so that no subscore can be quietly privileged after seeing the data. Any organisation whose status is defunct, dormant, inactive, ceased, acquired, merged or liquidated is forced to 0 and sorted to the bottom of the atlas.

1 · Clinical evidence

A base value from the strongest evidence design claimed, plus log-scaled volume credit, so that many small studies never outweigh one rigorous one.

Volume bonus: min(12, 6·log₂(1+trials)) + min(10, 4·log₂(1+publications)).

2 · Regulatory maturity

Two design commitments make this subscore comparable across the world. Jurisdiction neutrality: an FDA clearance, an EU MDR CE mark, a UKCA mark, a DiGA listing and a PMDA, TGA, Health Canada, ANVISA, NMPA, MFDS, CDSCO or SAHPRA authorisation count as equivalent attainment; no single regulator is privileged. Applicability: peer support, coaching, harm-reduction supply and B2B analytics need no device or drug approval, so they are scored against their own class ceiling rather than penalised against a ceiling they were never eligible for.

The record is first routed to a pathway, then scored as floor + attainment × (ceiling − floor):

Breadth bonus: at attainment ≥ 0.70, authorisation in 2 distinct jurisdictions adds +4 and 3 or more adds +8 — again without regard to which regulators they are.

3 · Commercial traction

Stage base — bootstrapped 38, grant-funded 34, angel 38, pre-seed 42, seed 52, Series A 64, Series B 74, Series C+ 84, public 90, acquired 78, non-profit 40, unknown 32 — plus min(15, 3.2·log₁₀(funding / 100 000)) so that a mega-round cannot dominate and a small non-US round is not flattened to zero. Revenue adds +16 (profitable), +10 (revenue-generating) or +3 (loss-making). A further +8 marks payer, insurer, NHS, Krankenkasse, Medicaid/Medicare, health-system or employer distribution — traction that venture funding alone never captures, and the usual shape of non-US and non-profit models.

4 · Innovation

Base 34, plus 5 per distinct novel-mechanism term matched across product types, delivery modes, innovation categories and tags (capped at +36), plus 3 per innovation category (capped +12), plus 2 per delivery mode (capped +8), plus 5 when the company covers both substance and behavioural addictions. The term list is explicit — AI/ML, digital therapeutics, biomarkers, genomics, VR/AR, wearables, implants, neurostimulation, remote monitoring, telehealth, MAT, harm reduction, peer support, precision medicine, gamification, chatbots, contingency management, just-in-time adaptive intervention, psychedelics, closed-loop and passive sensing.

Known biases of this rubric

• Text-derived: a company that documents its regulatory position badly scores low even if its position is strong.
• Disclosure-driven: undisclosed funding reads as no funding, which depresses non-US and non-profit records.
• Priority ≠ quality: the composite ranks what merits monitoring, not clinical merit or investability.

• It does not verify factual claims inside a page; only that the page exists and is about the right company.
• It does not score editorial bias.
• Funding figures, regulatory status, and trial counts still depend on the AI research step; they should be cross-checked before citation.

The preregistered hypotheses (OSF v1.0, §6.2) are computed live against the current atlas below. Per the registration, tests on data collected before the 2026-08-26 baseline are exploratory; confirmatory runs are scheduled at weeks 26 and 52 on the weekly snapshot series.

Computing…

If the rubric above is to be trusted, its output has to be inspectable. This panel is computed live over the active atlas: how each subscore is distributed, how strongly the subscores duplicate each other (a high ρ between two subscores means the composite is double-counting one signal), which subscore actually separates the top decile of the ranking, and whether regulatory scores differ by pathway once the pathway-relative ceilings are applied. Exploratory and descriptive throughout.

Computing…

A descriptive overlay, computed live: disclosed capital is compared against each category's share of the active atlas. A positive disparity means a substance category receives less capital than its company count would predict; the regional view contrasts capital share with organisation share. These are signals for further investigation, not causal estimates — disclosed funding is incomplete and skews Western.

Computing…

The Atlas is a descriptive monitoring instrument. It is strong evidence about structure — who exists, where, working on what, with what legitimacy and what capital — and weak evidence about cause. It cannot show that funding produced adoption, or that evidence produced commissioning. Read alongside other layers (registry data, commissioning records, interviews), it is one triangulated view of an innovation ecosystem, not an index of merit.

Legitimacy

Three legitimacies are tracked separately and never collapsed: regulatory(clearance, marking, listing), clinical (trial design and replication), and source authority (who says so). A company can hold market legitimacy with almost no clinical legitimacy, and the score structure is designed to make that visible rather than average it away.

Demand

Addiction is a field of large latent demand and weak demand activation: need vastly exceeds treatment-seeking, and treatment-seeking exceeds funded provision. So commercial traction here measures activated, payable demand — not need. Categories with heavy stigma and criminalisation (stimulants, opioids in many jurisdictions, gambling in some) systematically under-read on the commercial subscore for reasons that have nothing to do with product quality.

Capital

Four capital types behave differently and are recorded distinctly where disclosed: venture, patient capital (foundations, mission investors, long-horizon funds), public grant funding (NIH SBIR/STTR, Horizon Europe, national innovation agencies), and revenue from payers and public commissioners. Venture-only lenses read grant-funded and non-profit organisations as failures; adding NIH RePORTER and CORDIS to discovery is a direct correction to that.

Market fixing and market shaping

Some entries fix an existing market — cheaper, faster, better-targeted delivery of a service that already has a buyer. Others attempt to shape one: creating a category, a reimbursement route, or a standard where none existed (DiGA, prescription digital therapeutics, drug-checking services). The two face different risk and reward profiles, and the Atlas should not be read as ranking them on one axis.

The pathway, and where it breaks

Evidence → Pilot → Commissioning → Adoption → Scale

Most attrition in this field is not scientific failure but handoff failure between those stages — a pilot that never finds a commissioner, a cleared device with no reimbursement code. The constraints that produce those breaks are structural: stigma, misaligned incentives, fragmented commissioning, thin data infrastructure, workforce capacity, and trust and governance. Site status, successor records and dormancy flags exist so that attrition is recorded rather than quietly deleted.

Ecosystem, not a sector

The relevant system is wider than health technology: clinicians and providers, payers and commissioners, people with lived and living experience, families and communities, the third sector, housing, welfare, criminal justice, regulators, research and education, funders and policymakers. The Atlas indexes only the innovator layer. Treat gaps in it as questions about the surrounding system, not as verdicts on the companies in it.